User Modeling is a field of increasing importance for industrial applications, especially for information retrieval from large data-bases using browsing as a search strategy. The most of the research in this field, however, has been theoretical. We have implemented a new architecture for user modeling based on analysis of the tasks performed by the users. It allows adaptive browsing support for users with different levels of experience, data-protection, and a degree of adaptability according to the preferences of individual users. This architecture was applied in building a user modeling component for a hypermedia-based information system for hospital information which is now being experimented.